关键词: 多目标优化, 进化算法, 静态贝叶斯博弈

Abstract: Nowadays most of traditional Multi-Objective Evolutionary Algorithms（MOEAs） search global Pareto front based on the non-dominant sorting method. By this way evolutionary populations can get high convergence performance in the early iterative stage, however there maybe exists convergence stagnation for complex problems in the late iterative stage, which cuts off the global optimization ability. To address this problem a method based on Static Bayesian Game strategy Multi-Objective Evolutionary Algorithm（SBG-MOEA） is proposed. Each optimization object is emulated as one game participant and their convergence performances to Pareto solution set are mapped to the game incomes in the algorithm. The profit and loss Nash equilibrium game mechanism is applied to drive the evolutionary population chasing the Pareto front. Theoretical analysis proves that the approach can converge to the global Pareto optimal set. Compared with the classic multi-objective evolutionary algorithms, such as NAGAII, the simulation optimization results show that Bayesian game strategy can enhance the global optimal searching ability of MOEAs.

Key words: multi-objective optimization, evolutionary algorithm, static Bayesian game